Table 1.
Error rate comparison between RF, HSVM, Omni-PolyA and our model (DeeReCT-PolyA) on the Dragon human poly(A) data
| Variants | Size | Error Rate (%) |
||||
|---|---|---|---|---|---|---|
| RF | HSVM | Omni- PolyA | DeeReCT- PolyA | Rel | ||
| AATAAA | 5190 | 20.06 | 18.59 | 14.02 | 11.81 | 2.21 |
| ATTAAA | 2400 | 18.42 | 16.21 | 12.50 | 9.00 | 3.50 |
| AAAAAG | 1250 | 16.64 | 9.36 | 10.80 | 5.77 | 3.59 |
| AAGAAA | 1230 | 11.06 | 5.45 | 4.87 | 7.76 | −2.89 |
| TATAAA | 880 | 19.55 | 15.34 | 13.52 | 7.69 | 5.83 |
| AATACA | 780 | 19.36 | 11.15 | 13.85 | 10.45 | 0.70 |
| AGTAAA | 690 | 27.83 | 16.96 | 14.49 | 9.55 | 4.94 |
| ACTAAA | 670 | 22.09 | 14.33 | 13.13 | 10.72 | 2.41 |
| GATAAA | 460 | 20.00 | 9.57 | 8.48 | 8.04 | 0.44 |
| CATAAA | 410 | 18.54 | 9.27 | 13.41 | 9.02 | 0.25 |
| AATATA | 410 | 24.88 | 12.68 | 14.39 | 8.78 | 3.90 |
| AATAGA | 370 | 18.38 | 5.14 | 11.62 | 4.59 | 0.55 |
| Average | – | 19.19 | 14.42 | 12.43 | 9.57 | 2.86 |
Note: Rel denotes the improvement of DeeReCT-PolyA with respect to the best of the other three methods. Bold indicates the error rate of the best model for each PAS motif variant. Average is the weighted average of all motif variants with the size as weights. While results of all three previous methods are reported for 12 variant-specific models, the results of DeeReCT-PolyA are the performance of one single generic model that deals with all 12 variants.